Research on Energy-Saving Routing Technology Based on Deep Reinforcement Learning

With the vigorous development of Internet, network traffic data centers has exploded, and at same time, energy consumption also increased rapidly. Existing routing algorithms only realize optimization through Quality Service (QoS) Experience (QoE), which ignores center networks. Aiming this problem, paper proposes an Ee-Routing algorithm, is energy-saving algorithm based on deep reinforcement learning. First, our method takes performance plane in software-defined as joint goal establishes energy-efficient scheduling scheme for elephant flows mice flows. Then, we use Deep Deterministic Policy Gradient (DDPG), a learning framework, to achieve continuous goals. The training process Convolutional Neural Network (CNN), can effectively improve convergence efficiency algorithm. After converges, path weights are output, balanced completed. Finally, results show that good stability. Compared with DQN-EER improves saving percentage by 13.93%, compared EARS reduces delay 13.73%, increases throughput 10.91%, packet loss rate 13.51%.